Skip to content

Samsen Internal Framework

This is Samsen's internal framework content for AI-Assisted Design. It is read-only reference material.

The Shipping AI Designer — Core Thesis

The Concept

The "Shipping AI Designer" is a new category of practitioner: an expert product designer who ships production code using AI tools — specifically Claude Code and Figma MCP — without becoming a traditional developer.

This is not about designers learning to code. It's about designers describing what they want, reviewing it in the browser, and describing adjustments. The same workflow as collaborating with a developer, but faster and with more direct control.

Why This Matters Now

The tools have crossed a threshold. Claude Code with MCP integrations can read a Figma file, understand a design system's token architecture, generate production-quality components, and iterate based on natural language feedback. The gap between design intent and shipped code has collapsed.

But the industry conversation is stuck in two outdated frames:

  1. "Designers should learn to code" — This has been debated for 15 years and the answer is still no. Designers should design. The value isn't in typing syntax.
  2. "AI will replace designers" — This misunderstands what designers do. The craft is in decisions, not deliverables. AI handles the production; designers direct the outcome.

The Shipping AI Designer sidesteps both. Designers keep their craft expertise and gain the ability to ship directly, without a handoff and without writing code.

The Five Principles

1. Hybrid Workflow

Figma for visual exploration ("drawing to think"), Claude Code for implementation and iteration. These aren't competing tools — they serve different cognitive modes. Figma is where you discover what something should be. Claude Code is where you make it real.

See: hybrid-workflow.md

2. Design System Alignment Is the Bridge

For AI to generate reliable output, the design system tokens in Figma must match the CSS/code tokens exactly. Figma variables ↔ CSS custom properties. This alignment is what makes AI-assisted implementation predictable rather than random.

See: ../design-systems/token-architecture.md

3. The 80/60 Insight

80% of design system work is maintenance. 60% of product designer craft time goes to iteration, QA, and file organization. These are exactly the tasks AI handles best. The opportunity isn't replacing creative work — it's eliminating the drudgery that surrounds it.

See: the-80-60-insight.md

4. The Last Mile Problem

The real challenge isn't generating code. Every AI tool can produce a component. The hard part is shipping that component to a production codebase: matching existing patterns, respecting architectural conventions, passing CI, working in the real dependency tree. This is what separates demos from production.

See: ../design-systems/last-mile-problem.md

5. Designers Don't Write Code

They describe what they want. They review in the browser. They describe adjustments. This is the same feedback loop designers already use with developers — but with AI, the cycle time drops from days to minutes. The designer's vocabulary doesn't change. Their judgment doesn't change. Only their leverage changes.

The Practitioner Profile

A Shipping AI Designer:

  • Has deep product design expertise (this is a prerequisite, not something the framework teaches)
  • Understands their design system at the token level (variables, naming conventions, semantic structure)
  • Can describe implementation requirements in natural language with enough specificity for AI to act on
  • Reviews output in the browser, not in generated code
  • Knows enough about git to manage branches and PRs (but doesn't need to understand the code in them)
  • Ships to production — not just prototypes, not just demos

What This Is Not

  • Not a no-code tool. Code is generated, it's real, it ships. The designer just doesn't write it.
  • Not for beginners. You need design expertise to direct AI effectively. Garbage in, garbage out.
  • Not replacing developers. Complex application logic, data architecture, performance optimization — these still need engineers. The Shipping AI Designer handles the UI/component layer.
  • Not Figma-to-code export. Those tools generate static snapshots. This is an iterative, conversational workflow with AI that understands context.